We study the jet kinematics of the blazar S5 0716+714 during its active state in 2003-2004 with multi-epoch VLBI observations. Aims. We present a kinematic analysis of the large-scale (0-12 mas) jet of 0716+714, based on the results of six epochs of VLBA monitoring at 5 GHz. Additionally, we compare kinematic results obtained with two imaging methods based on different deconvolution algorithms. The blazar 0716+714 has a diffuse large-scale jet and a very faint bright compact core. Experiments with simulated data showed that the conventional data reduction procedure based on the CLEAN deconvolution algorithm does not perform well in restoring this type of structure. This might be the reason why previous kinematic studies of this source yielded ambiguous results. In order to obtain accurate kinematics of this source, we independently applied two imaging techniques to the raw data: the conventional method, based on difference mapping, which uses CLEAN deconvolution, and the generalized maximum entropy method (GMEM) realized in the VLBImager package developed at the Pulkovo Observatory in Russia. The results of both methods give us a consistent kinematic scenario: the large-scale jet of 0716+714 is diffuse and stationary. Differences between the inner (0-1 mas) and outer (1-12 mas) regions of the jet in brightness and velocity of the components could be explained by the bending of the jet, which causes the angle between the jet direction and the line of sight to change from ~5 deg to ~11 deg. For the source 0716+714 both methods worked at the limit of their capability.

We observed the blazar \object{0716+714} with the VLBA during its active state in 2003-2004. In this paper we discuss multi-frequency analysis of the inner jet (first 1 mas) kinematics. The unprecedentedly dense time sampling allows us to trace jet components without misidentification and to calculate the component speeds with good accuracy. In the smooth superluminal jet we were able to identify and track three components over time moving outwards with relatively high apparent superluminal speeds (8.5-19.4 $c$), which contradicts the hypothesis of a stationary oscillating jet in this source. Component ejections occur at a relatively high rate (once in two months), and they are accompanied by mm-continuum outbursts. Superluminal jet components move along wiggling trajectories, which is an indication of actual helical motion. Fast proper motion and rapid decay of the components suggest that this source should be observed with the VLBI at a rate of at least once in one or two months in order to trace superluminal jet components without confusion.